Skip to contentSkip to conversation

btrfs: change the set_page_extent_mapped() call into an ASSERT()

PATCH In review Qu Wenruo 2 participants 3 messages Jul 27, 2021 Btrfs file system

lore
Share & tools
lore

Qu Wenruo

author
#0 lore raw

PATCH 1/1 btrfs: change the set_page_extent_mapped() call into an ASSERT()

Btrfs uses set_page_extent_mapped() to properly setup a page.

That function would set PagePrivate, and populate needed structure for subpage.
The timing of calling set_page_extent_mapped() happens before reading a page or dirtying a page.
Thus when we got a page to write back, if it doesn't have PagePrivate, it is a big problem in code logic.

Calling set_page_extent_mapped() for such page would just mask the problem.
Furthermore, for subpage case, we call subpage error helper to clear the page error bit before calling set_page_extent_mapped().
If we really got a page without Private bit, it can call kernel NULL pointer dereference.

So change the set_page_extent_mapped() call to an ASSERT(), and move the check before any page status update call.

---
 fs/btrfs/extent_io.c | 12 ++++++------
 1 file changed, 6 insertions(+), 6 deletions(-)
fs/btrfs/extent_io.c +6 -6
index 62f0ed2de2b9..219add264acf 100644
@@ -4099,6 +4099,12 @@ static int __extent_writepage(struct page *page, struct writeback_control *wbc,
 
 	WARN_ON(!PageLocked(page));
 
+	/*
+	 * All dirty page to be written should have PagePrivate set by
+	 * set_page_extent_mapped() when creating the page.
+	 */
+	ASSERT(PagePrivate(page));
+
 	btrfs_page_clear_error(btrfs_sb(inode->i_sb), page,
 			       page_offset(page), PAGE_SIZE);
 
@@ -4115,12 +4121,6 @@ static int __extent_writepage(struct page *page, struct writeback_control *wbc,
 		flush_dcache_page(page);
 	}
 
-	ret = set_page_extent_mapped(page);
-	if (ret < 0) {
-		SetPageError(page);
-		goto done;
-	}
-
 	if (!epd->extent_locked) {
 		ret = writepage_delalloc(BTRFS_I(inode), page, wbc,
 					 &nr_written);
Signature · 1 line
2.32.0
author
#1 lore raw
Quoted 16 lines · Qu Wenruo

Btrfs uses set_page_extent_mapped() to properly setup a page.

That function would set PagePrivate, and populate needed structure for subpage.
The timing of calling set_page_extent_mapped() happens before reading a page or dirtying a page.
Thus when we got a page to write back, if it doesn't have PagePrivate, it is a big problem in code logic.

Calling set_page_extent_mapped() for such page would just mask the problem.
Furthermore, for subpage case, we call subpage error helper to clear the page error bit before calling set_page_extent_mapped().
If we really got a page without Private bit, it can call kernel NULL pointer dereference.

Quoted 4 lines

So change the set_page_extent_mapped() call to an ASSERT(), and move the check before any page status update call.

Please discard the patch.

Although I haven't hit any problem testing the patch, it's still possible that we have a special page that would need cow fixup.

Such page will be:

- Dirty

- Not Private
Thus no page->private, this could still cause problem for subpage case though

- No EXTENT_DELALLOC flags set for any range inside the page Thus writepage_delalloc() will not find a delalloc range inside the page.

Such page will be caught by btrfs_writepage_cow_fixup(), but it will trigger the ASSERT() added by this patch.

Thanks,
Qu

Quoted 35 lines
---
  fs/btrfs/extent_io.c | 12 ++++++------
  1 file changed, 6 insertions(+), 6 deletions(-)
fs/btrfs/extent_io.c +6 -6
index 62f0ed2de2b9..219add264acf 100644
@@ -4099,6 +4099,12 @@ static int __extent_writepage(struct page *page, struct writeback_control *wbc,
 
  	WARN_ON(!PageLocked(page));
 
+	/*
+	 * All dirty page to be written should have PagePrivate set by
+	 * set_page_extent_mapped() when creating the page.
+	 */
+	ASSERT(PagePrivate(page));
+
  	btrfs_page_clear_error(btrfs_sb(inode->i_sb), page,
  			       page_offset(page), PAGE_SIZE);
 
@@ -4115,12 +4121,6 @@ static int __extent_writepage(struct page *page, struct writeback_control *wbc,
  		flush_dcache_page(page);
  	}
 
-	ret = set_page_extent_mapped(page);
-	if (ret < 0) {
-		SetPageError(page);
-		goto done;
-	}
-
  	if (!epd->extent_locked) {
  		ret = writepage_delalloc(BTRFS_I(inode), page, wbc,
  					 &nr_written);
#2 lore raw
Quoted 37 lines · Qu Wenruo
Quoted 16 lines · Qu Wenruo

Btrfs uses set_page_extent_mapped() to properly setup a page.

That function would set PagePrivate, and populate needed structure for subpage.
The timing of calling set_page_extent_mapped() happens before reading a page or dirtying a page.
Thus when we got a page to write back, if it doesn't have PagePrivate, it is a big problem in code logic.

Calling set_page_extent_mapped() for such page would just mask the problem.
Furthermore, for subpage case, we call subpage error helper to clear the page error bit before calling set_page_extent_mapped().
If we really got a page without Private bit, it can call kernel NULL pointer dereference.

Quoted 4 lines

So change the set_page_extent_mapped() call to an ASSERT(), and move the check before any page status update call.

Please discard the patch.

Although I haven't hit any problem testing the patch, it's still possible that we have a special page that would need cow fixup.

Such page will be:

- Dirty

- Not Private
Thus no page->private, this could still cause problem for subpage case though

Quoted 6 lines

- No EXTENT_DELALLOC flags set for any range inside the page Thus writepage_delalloc() will not find a delalloc range inside the page.

Such page will be caught by btrfs_writepage_cow_fixup(), but it will trigger the ASSERT() added by this patch.

Hm yeah the mismatch between dirty/private/delalloc sounds exactly like work for the cow fixup.